Everyone knows that I think 23andme provides a great service. But I’ve had some criticisms in the past. Several years ago I thought it was rather strange of them to limit their chromosome painting feature to only a few ancestral components when it produced strange confusing results (e.g., many East Africans being mostly European in ancestry). Over the past few years they’ve nicely expanded their ancestry components, addressing this concern. But at this point my own inclination is to say that they’ve gone too far. For example a friend who is ethnically Japanese from Honshu gets these results: 76.1% – Japanese, 3.5% – Korean, 19.7% – Nonspecific East Asian, 0.4% – Nonspecific East Asian & Native American and 0.3% – European. I can give you reasonable explanations for these proportions, but it’s going to be confusing for many Japanese to be told they’re only 3/4 Japanese genetically. The issue is the scope of their reference population. It’s not capturing the diversity of the whole population of Japan.
But this is a minor concern in comparison to something else I’ve noticed. Here are some ancestral proportions from my family:
|French & German||British & Irish|
I’ll tell you right now that I don’t have any European ancestry according to 23andMe, so my daughter’s elevated French & German and British & Irish have nothing to do with her paternal lineage. In addition I can look at the chromosome level ancestry. Most of my daughter’s chromosome 3 and all of chromosome 6 are French & German (at least the ones obviously inherited from her mother). But there is no French & German ancestry for my wife on these chromosomes (either copy). There is one recombination event from her maternal chromosome 6 and three for chromosome 3, but I don’t think that should have such a large effect.
So hypotheses? My own hunch is that clusters like French & German are somewhat artificial, insofar as they cover a very large geographic area (though granted Europe from the Bay of Biscay to the Elbe is definitely relatively genetically homogeneous). People of mixed European ancestry, like many American whites, often may resolve strangely because the methods used have a difficult time distinguishing mixed ancestry from populations which are composed of mixed ancestry (like many American whites the “French” have diverse ancestries from different regions of Europe, so many Americans may look somewhat “French”). A friend from Guatemala who is ethnically mestizo of many generations has 20% unassigned ancestry, presumably because so many recombination events have intercalated Amerindian and European segments, making it impossible for 23andMe to give a correct assignment. My parents and myself have unassigned proportions of nearly 10%, likely due to ancient admixture between our South and East Asian components, which dates back over 1,000 years in the past.
Personal genomics services are great, and I heartily recommend them. But people should be careful about taking all the results at face value. It’s like with anything else, be an informed and cautious consumer. It can be great for some things, like Dan MacArthur’s South Asian ancestry, which literally jumped out of the genetic background. But the finer you get in the grain, the more confusion will result.
Update: In the comments 23andMe scientist Eric Durand asked if I’d enabled split view. The reason he asks is that 23andMe phases the genotypes (reconstructed each physically linked segment of chromosome of the pair) before assigning ancestry along a segment. If you don’t have family information you have to use population based information. But more powerful are parental trios, since offspring simply have mixed & matched segments of their parents. And yes, I do have split view enabled, and checked it. Same weird result: